Peter Cole

Wifi-based localisation systems have gained significant interest with many researchers proposing different localisation techniques using publicly available datasets. However, these datasets are limited because they only contain Wifi fingerprints collected and labelled by users, and they are restricted to indoor locations. We have generated the first Wifi-based localisation datasets for a GPS-denied open area. We selected a busy open area at Murdoch University to generate the datasets using so-called “smart bins”, which are rubbish bins that we enabled to work as access points. The data gathered consists of two different datasets. In the first, four users generated labelled WiFi fingerprints for all available Reference Points using four different smartphones. The second dataset includes 2450865 auto-generated rows received from more than 1000 devices. We have developed a light-weight algorithm to label the second dataset from the first and we proposed a localisation approach that converts the second dataset from asynchronous format to synchronous, applies feature engineering and a deep learning classifier. Finally, we have demonstrated via simulations that by using this approach we achieve higher prediction accuracy, with up to 19% average improvement, compared with using only the fingerprint dataset.

In recent years, Wifi-based localisation systems have gained significant interest because of the lack of Global Positioning System (GPS) signal in indoor and certain open areas. Over the past decade, many datasets have been introduced to enable researchers to compare different localisation techniques. Existing datasets, however, have failed to cover open areas such as parks in cases where GPS is still unavailable, and there is a lack of Wifi access points. Also, the existing datasets only focus on getting Wifi fingerprint collected and labelled by users. To the best of our knowledge, no dataset provides Received Signal Strengths (RSS) collected by Wireless Access Points (APs). In this work, we offer two datasets publicly. The first is the Fingerprint dataset in which four users generated 16,032 accurate and consistently labelled WiFi fingerprints for all available Reference Points (RPs) in a central and busy area of Murdoch University, known as Bush Court. The second is the APs dataset that includes 2,450,865 auto-generated records received from 1000 users’ devices, including the four users, associated with Wifi signal strengths. To overcome the Wifi coverage problem for the Bush Court, we attached our previously designed Wireless Sensor Nodes (WSNs) to existing garbage bins, enabling them to provide real-time environmental sensing and act as soft APs that sense MAC addresses and Wifi signals from surrounding devices.

Recently, street furniture, including bins, seats, and bus shelters, has become smart as it has been equipped with environmental sensors, wireless modules, processors, and microcontrollers. Accordingly, smart furniture is expected to become a vital part of the IoT infrastructure and one of the drivers of future smart cities. This work focuses on how smart street furniture can be exploited within the IoT architecture as a basis of recommender systems, toward achieving smart cities' different components. We present and discuss recent relevant work as well as the key challenges and opportunities for future research. We explain that much work is still required when it comes to combining scalability, real-time processing, smart furniture, and recommender systems.

This paper presents the Weighted Random Early Detection (WTRED) strategy for congestion handling in TCP networks. WTRED provides an adjustable weight parameter to increase the sensitivity of the average queue size in RED gateways to the changes in the actual queue size. This modification, over the original RED proposal, helps gateways minimize the mismatch between average and actual queue sizes in router buffers. WTRED is compared with RED and FRED strategies using the NS-2 simulator. The results suggest that WTRED outperforms RED and FRED. Network performance has been measured using throughput, link utilization, packet loss and delay.

This work presents the Risk Threshold RED (RTRED) congestion control strategy for TCP networks. In addition to the maximum and minimum thresholds in existing RED-based strategies, we add a third dropping level. This new dropping level is the risk threshold which works with the actual and average queue sizes to detect the immediate congestion in gateways. Congestion reaction by RTRED is on time. The reaction to congestion is neither too early, to avoid unfair packet losses, nor too late to avoid packet dropping from time-outs. We compared our novel strategy with RED and ARED strategies for TCP congestion handling using a NS-2 simulation script. We found that the RTRED strategy outperformed RED and ARED.

This work presents the Weighted Random Early Detection (WTRED) strategy for congestion handling in TCP networks. The strategy dynamically adjusts RED's maximum threshold, minimum threshold and weight parameters to increase network performance. This work describes RED and FRED implementations and highlights their disadvantages. Using the NS-2 simulator, we compare WTRED with these classic congestion control strategies. The simulation results demonstrate the shortcomings of RED and FRED. The results also show that WTRED achieves greater link utilization and throughput than RED and FRED.

Network congestion is a phenomenon caused by the extreme demand of restricted network resources. Various congestion control strategies have been proposed to increase network performance. This study suggests that there is a mismatch between the microscopic and macroscopic behavior in (Random Early Detection) RED’s queue management mechanism. This work investigates this problem and propose QSRED (Queue Sectors RED) to avoid unsatisfactory performance. QSRED is simulated against RED and ERED (Effective RED) by measuring: throughput, link utilization, packets loss and average delay using the NS2 simulator. The results suggest that Queue Sectors RED (QSRED) helps RED overcome the mismatch between microscopic and macroscopic behavior of queue length dynamics.

Computer vision systems have varied applications and often the first goal of a vision system is the recognition of 3-D rigid objects. Model-based recognition systems rely upon a model to represent an object both tersely and uniquely to allow for efficient matching. The View Signature II (VSII) representation is a viewer-centred modelling scheme. The VSII representation requires one VSII view signature to represent a single view of an object and is invariant to rotation about the z axis. This is achieved by constructing a linear string of alpha-numeric characters from circular strings representing different concentric levels in a view of an object. A view signature is an ordered list of junction and level signatures and can be considered a string for matching purposes. Junction signatures represent the cotermination of edges and contours with their respective adjacent surfaces. A feature of view signature construction is the uniform processing of both the stored model and matching input, thereby removing the possibility of incompatibility between model and input. The motivation for the View Signature representation, its syntax and a technique for modelling 3-D rigid objects is described.

The View Signature II (VSII) representation is a viewer-centred modelling scheme. The VSII representation requires one VSII view signature to represent a single view of an object and is invariant to rotation about the z axis. This is achieved by constructing a linear string of alpha-numeric characters from circular strings representing different concentric levels in a view of an object. The tokens and algorithmic methods for the construction of a VSII produce a language for the modelling of 3-D rigid solid objects. This paper briefly describes the VSII representation and the general construction methods. Due to the cyclic nature of the representation it has been possible to define a taxonomy of occlusion in VSII representations. This paper discusses the differing forms of occlusion and gives methods where appropriate to assist in matching occluded views with complete views in the model store.